5 Questions to Ask Before Connecting AI to Company Email

Buying an AI tool is easy. Connecting AI to company email and explaining it to your security officer six months later is not. These five questions take about an hour to ask and answer, and they decide whether the tool is something you can defend in an audit, a customer review or a bad week.

In short

Ask where the data goes, how access is granted, what you can limit, what is recorded and what happens on exit.
Get every answer in writing. A vague answer is an answer.
If a vendor cannot answer three of the five clearly, pause the rollout.

Why a checklist beats a demo

A demo shows what an AI assistant can do. It rarely shows what happens to your mail while it does it. Email contains customer data, contracts, price lists, HR matters and personal messages, so the questions that matter are about control, not cleverness.

Use the same five questions with every vendor, including us. Put the answers side by side and the differences become obvious in minutes.

Five questions, asked of every vendor in the same order.

1. Where does my data go?

Ask which AI model sees the text, in which region it runs and whether the vendor stores the content. The answer depends on the setup, so ask about each one you are offered.

With a tool you already pay for, such as Claude or ChatGPT, the text goes to that provider under your own account and terms. With a vendor-hosted chat window, it goes to the model the vendor chose. In AddSync AI these are option A and option B, and in option B the model runs in an EU region.

A good answer	A warning sign
Names the model, the region and the storage rule, and offers it in a contract or policy.	“We take privacy very seriously.” Nothing about the model or the region.

2. How is access given?

The weakest design asks every employee to paste a mailbox password into a third-party tool. Passwords leak, people leave, and nobody can say later what the tool did with them.

Look for access that IT creates and can switch off in one place. For Exchange on premises that is a dedicated user with mailbox permissions. For Microsoft 365, Google and Outlook.com it is an approved sign-in, not a shared password.

3. Can I limit what the AI sees?

You need rules you control, not a promise that the AI will behave. The useful controls are simple: exclude folders, exclude senders and domains, exclude words or patterns, and set an age limit for messages and attachments.

A rule that is set once for the whole company is easier to explain to an auditor than a different setting for every person. Ask whether the vendor’s filters work that way, and test one before you sign.

4. Is there a record?

You should be able to see who used the AI, when, and what it did. Without a journal you cannot investigate an incident or show a regulator that you had control. In AddSync AI the journal keeps one line per tool: who, when and which button, without the text of the mail.

Also ask who can read the journal. A record that only the vendor can see helps the vendor, not you.

5. What happens when we leave?

Ask how access is revoked, how fast, and what is deleted. The best answer is short and specific. If you hear “it depends”, ask what it depends on and write that down.

How to score the answers

Give each question one point for a clear, written answer and zero for anything else. Three out of five is the minimum to continue. Four or five means you can move to a small pilot.

ScoreAdd up the points for each vendor, not for the best demo.
ShareSend the answers to the security officer and the data protection contact before the pilot starts.
RepeatAsk the five questions again after the pilot, with real journal entries in hand.

Our own answers are published on the trust page, in the security and compliance overview and in the privacy policy.

What good answers sound like

Vendors rarely lie. They answer a different question than the one you asked. Keep your own wording in front of you and compare.

Question	A specific answer	A vague answer
Where does my data go?	“Option A: your own Claude or ChatGPT account. Option B: our chat window, model in an EU region.”	“It is processed securely in the cloud.”
How is access given?	“One global user with mailbox permissions. No impersonation.”	“Users connect their accounts.”
Can I limit what the AI sees?	“Four filter types, set once for the whole company.”	“The AI only uses what it needs.”
Is there a record?	“One journal line per tool: who, when and which button. No mail text.”	“We keep logs.”
What happens when we leave?	“Switch AI off for one person, a group or everyone. Mail was never moved.”	“Contact support.”

The examples in the middle column describe how AddSync AI works today. Check them against the trust page, and ask every other vendor for the same level of detail.

Questions and answers

Who should ask these questions?

The IT lead and the person responsible for data protection or security, with a business owner in the room. Ten minutes of the business owner’s time saves weeks of arguments later.

Are five questions enough?

They cover the decisions that are hard to undo. Contract terms, price and support are separate topics, and a pilot is the best way to test the rest.

Does this apply to tools my staff already use?

Yes. If people use an AI tool on their own, ask the same questions about it, and consider offering an approved alternative so mail stays under your rules.

Where can I read AddSync's answers?

On the trust page and in the security and compliance overview. Both are linked in this article.

Ask us the five questions

Start a 14-day pilot and put our answers next to the ones you get from other vendors.

Start the 14-day pilot See plans

Keep reading

AddSync AI trust: How data is handled, in plain language.
Security and compliance: What the AI can reach and what you can prove.
AddSync AI vs ChatGPT and Claude on their own: What changes when IT adds a guard.
What is Mail AI?: The idea in five minutes.

Written by the AddSync team.

Get new guides by email

One email when we publish something worth reading about AddSync AI and AddSync Calendar.